1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* AF_RXRPC sendmsg() implementation.
4 * Copyright (C) 2007, 2016 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10 #include <linux/net.h>
11 #include <linux/gfp.h>
12 #include <linux/skbuff.h>
13 #include <linux/export.h>
14 #include <linux/sched/signal.h>
17 #include <net/af_rxrpc.h>
18 #include "ar-internal.h"
21 * Return true if there's sufficient Tx queue space.
23 static bool rxrpc_check_tx_space(struct rxrpc_call *call, rxrpc_seq_t *_tx_win)
26 *_tx_win = call->tx_bottom;
27 return call->tx_prepared - call->tx_bottom < 256;
31 * Wait for space to appear in the Tx queue or a signal to occur.
33 static int rxrpc_wait_for_tx_window_intr(struct rxrpc_sock *rx,
34 struct rxrpc_call *call,
38 set_current_state(TASK_INTERRUPTIBLE);
39 if (rxrpc_check_tx_space(call, NULL))
42 if (call->state >= RXRPC_CALL_COMPLETE)
45 if (signal_pending(current))
46 return sock_intr_errno(*timeo);
48 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
49 *timeo = schedule_timeout(*timeo);
54 * Wait for space to appear in the Tx queue uninterruptibly, but with
55 * a timeout of 2*RTT if no progress was made and a signal occurred.
57 static int rxrpc_wait_for_tx_window_waitall(struct rxrpc_sock *rx,
58 struct rxrpc_call *call)
60 rxrpc_seq_t tx_start, tx_win;
61 signed long rtt, timeout;
63 rtt = READ_ONCE(call->peer->srtt_us) >> 3;
64 rtt = usecs_to_jiffies(rtt) * 2;
69 tx_start = smp_load_acquire(&call->acks_hard_ack);
72 set_current_state(TASK_UNINTERRUPTIBLE);
74 if (rxrpc_check_tx_space(call, &tx_win))
77 if (call->state >= RXRPC_CALL_COMPLETE)
81 tx_win == tx_start && signal_pending(current))
84 if (tx_win != tx_start) {
89 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
90 timeout = schedule_timeout(timeout);
95 * Wait for space to appear in the Tx queue uninterruptibly.
97 static int rxrpc_wait_for_tx_window_nonintr(struct rxrpc_sock *rx,
98 struct rxrpc_call *call,
102 set_current_state(TASK_UNINTERRUPTIBLE);
103 if (rxrpc_check_tx_space(call, NULL))
106 if (call->state >= RXRPC_CALL_COMPLETE)
109 trace_rxrpc_txqueue(call, rxrpc_txqueue_wait);
110 *timeo = schedule_timeout(*timeo);
115 * wait for space to appear in the transmit/ACK window
116 * - caller holds the socket locked
118 static int rxrpc_wait_for_tx_window(struct rxrpc_sock *rx,
119 struct rxrpc_call *call,
123 DECLARE_WAITQUEUE(myself, current);
126 _enter(",{%u,%u,%u,%u}",
127 call->tx_bottom, call->acks_hard_ack, call->tx_top, call->tx_winsize);
129 add_wait_queue(&call->waitq, &myself);
131 switch (call->interruptibility) {
132 case RXRPC_INTERRUPTIBLE:
134 ret = rxrpc_wait_for_tx_window_waitall(rx, call);
136 ret = rxrpc_wait_for_tx_window_intr(rx, call, timeo);
138 case RXRPC_PREINTERRUPTIBLE:
139 case RXRPC_UNINTERRUPTIBLE:
141 ret = rxrpc_wait_for_tx_window_nonintr(rx, call, timeo);
145 remove_wait_queue(&call->waitq, &myself);
146 set_current_state(TASK_RUNNING);
147 _leave(" = %d", ret);
152 * Notify the owner of the call that the transmit phase is ended and the last
153 * packet has been queued.
155 static void rxrpc_notify_end_tx(struct rxrpc_sock *rx, struct rxrpc_call *call,
156 rxrpc_notify_end_tx_t notify_end_tx)
159 notify_end_tx(&rx->sk, call, call->user_call_ID);
163 * Queue a DATA packet for transmission, set the resend timeout and send
164 * the packet immediately. Returns the error from rxrpc_send_data_packet()
165 * in case the caller wants to do something with it.
167 static void rxrpc_queue_packet(struct rxrpc_sock *rx, struct rxrpc_call *call,
168 struct rxrpc_txbuf *txb,
169 rxrpc_notify_end_tx_t notify_end_tx)
172 rxrpc_seq_t seq = txb->seq;
173 bool last = test_bit(RXRPC_TXBUF_LAST, &txb->flags);
175 rxrpc_inc_stat(call->rxnet, stat_tx_data);
177 ASSERTCMP(txb->seq, ==, call->tx_prepared + 1);
179 /* We have to set the timestamp before queueing as the retransmit
180 * algorithm can see the packet as soon as we queue it.
182 txb->last_sent = ktime_get_real();
185 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue_last);
187 trace_rxrpc_txqueue(call, rxrpc_txqueue_queue);
189 /* Add the packet to the call's output buffer */
190 spin_lock(&call->tx_lock);
191 list_add_tail(&txb->call_link, &call->tx_sendmsg);
192 call->tx_prepared = seq;
193 spin_unlock(&call->tx_lock);
195 if (last || call->state == RXRPC_CALL_SERVER_ACK_REQUEST) {
196 _debug("________awaiting reply/ACK__________");
197 write_lock_bh(&call->state_lock);
198 switch (call->state) {
199 case RXRPC_CALL_CLIENT_SEND_REQUEST:
200 call->state = RXRPC_CALL_CLIENT_AWAIT_REPLY;
201 rxrpc_notify_end_tx(rx, call, notify_end_tx);
203 case RXRPC_CALL_SERVER_ACK_REQUEST:
204 call->state = RXRPC_CALL_SERVER_SEND_REPLY;
206 WRITE_ONCE(call->delay_ack_at, now + MAX_JIFFY_OFFSET);
207 if (call->ackr_reason == RXRPC_ACK_DELAY)
208 call->ackr_reason = 0;
209 trace_rxrpc_timer(call, rxrpc_timer_init_for_send_reply, now);
213 case RXRPC_CALL_SERVER_SEND_REPLY:
214 call->state = RXRPC_CALL_SERVER_AWAIT_ACK;
215 rxrpc_notify_end_tx(rx, call, notify_end_tx);
220 write_unlock_bh(&call->state_lock);
224 /* Stick the packet on the crypto queue or the transmission queue as
227 rxrpc_queue_call(call, rxrpc_call_queue_tx_data);
231 * send data through a socket
232 * - must be called in process context
233 * - The caller holds the call user access mutex, but not the socket lock.
235 static int rxrpc_send_data(struct rxrpc_sock *rx,
236 struct rxrpc_call *call,
237 struct msghdr *msg, size_t len,
238 rxrpc_notify_end_tx_t notify_end_tx,
241 struct rxrpc_txbuf *txb;
242 struct sock *sk = &rx->sk;
243 enum rxrpc_call_state state;
245 bool more = msg->msg_flags & MSG_MORE;
248 timeo = sock_sndtimeo(sk, msg->msg_flags & MSG_DONTWAIT);
250 /* this should be in poll */
251 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
255 if (sk->sk_shutdown & SEND_SHUTDOWN)
257 state = READ_ONCE(call->state);
259 if (state >= RXRPC_CALL_COMPLETE)
262 if (state != RXRPC_CALL_CLIENT_SEND_REQUEST &&
263 state != RXRPC_CALL_SERVER_ACK_REQUEST &&
264 state != RXRPC_CALL_SERVER_SEND_REPLY)
268 if (call->tx_total_len != -1) {
269 if (len - copied > call->tx_total_len)
271 if (!more && len - copied != call->tx_total_len)
275 txb = call->tx_pending;
276 call->tx_pending = NULL;
278 rxrpc_see_txbuf(txb, rxrpc_txbuf_see_send_more);
281 rxrpc_transmit_ack_packets(call->peer->local);
284 size_t remain, bufsize, chunk, offset;
288 if (!rxrpc_check_tx_space(call, NULL))
291 /* Work out the maximum size of a packet. Assume that
292 * the security header is going to be in the padded
293 * region (enc blocksize), but the trailer is not.
295 remain = more ? INT_MAX : msg_data_left(msg);
296 ret = call->conn->security->how_much_data(call, remain,
297 &bufsize, &chunk, &offset);
301 _debug("SIZE: %zu/%zu @%zu", chunk, bufsize, offset);
303 /* create a buffer that we can retain until it's ACK'd */
305 txb = rxrpc_alloc_txbuf(call, RXRPC_PACKET_TYPE_DATA,
310 txb->offset = offset;
311 txb->space -= offset;
312 txb->space = min_t(size_t, chunk, txb->space);
317 /* append next segment of data to the current buffer */
318 if (msg_data_left(msg) > 0) {
319 size_t copy = min_t(size_t, txb->space, msg_data_left(msg));
321 _debug("add %zu", copy);
322 if (!copy_from_iter_full(txb->data + txb->offset, copy,
330 if (call->tx_total_len != -1)
331 call->tx_total_len -= copy;
334 /* check for the far side aborting the call or a network error
336 if (call->state == RXRPC_CALL_COMPLETE)
337 goto call_terminated;
339 /* add the packet to the send queue if it's now full */
341 (msg_data_left(msg) == 0 && !more)) {
342 if (msg_data_left(msg) == 0 && !more) {
343 txb->wire.flags |= RXRPC_LAST_PACKET;
344 __set_bit(RXRPC_TXBUF_LAST, &txb->flags);
346 else if (call->tx_top - call->acks_hard_ack <
348 txb->wire.flags |= RXRPC_MORE_PACKETS;
350 ret = call->security->secure_packet(call, txb);
354 rxrpc_queue_packet(rx, call, txb, notify_end_tx);
357 } while (msg_data_left(msg) > 0);
361 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE) {
362 read_lock_bh(&call->state_lock);
365 read_unlock_bh(&call->state_lock);
368 call->tx_pending = txb;
369 _leave(" = %d", ret);
373 rxrpc_put_txbuf(txb, rxrpc_txbuf_put_send_aborted);
374 _leave(" = %d", call->error);
388 if (msg->msg_flags & MSG_DONTWAIT)
390 mutex_unlock(&call->user_mutex);
391 *_dropped_lock = true;
392 ret = rxrpc_wait_for_tx_window(rx, call, &timeo,
393 msg->msg_flags & MSG_WAITALL);
396 if (call->interruptibility == RXRPC_INTERRUPTIBLE) {
397 if (mutex_lock_interruptible(&call->user_mutex) < 0) {
398 ret = sock_intr_errno(timeo);
402 mutex_lock(&call->user_mutex);
404 *_dropped_lock = false;
409 * extract control messages from the sendmsg() control buffer
411 static int rxrpc_sendmsg_cmsg(struct msghdr *msg, struct rxrpc_send_params *p)
413 struct cmsghdr *cmsg;
414 bool got_user_ID = false;
417 if (msg->msg_controllen == 0)
420 for_each_cmsghdr(cmsg, msg) {
421 if (!CMSG_OK(msg, cmsg))
424 len = cmsg->cmsg_len - sizeof(struct cmsghdr);
425 _debug("CMSG %d, %d, %d",
426 cmsg->cmsg_level, cmsg->cmsg_type, len);
428 if (cmsg->cmsg_level != SOL_RXRPC)
431 switch (cmsg->cmsg_type) {
432 case RXRPC_USER_CALL_ID:
433 if (msg->msg_flags & MSG_CMSG_COMPAT) {
434 if (len != sizeof(u32))
436 p->call.user_call_ID = *(u32 *)CMSG_DATA(cmsg);
438 if (len != sizeof(unsigned long))
440 p->call.user_call_ID = *(unsigned long *)
447 if (p->command != RXRPC_CMD_SEND_DATA)
449 p->command = RXRPC_CMD_SEND_ABORT;
450 if (len != sizeof(p->abort_code))
452 p->abort_code = *(unsigned int *)CMSG_DATA(cmsg);
453 if (p->abort_code == 0)
457 case RXRPC_CHARGE_ACCEPT:
458 if (p->command != RXRPC_CMD_SEND_DATA)
460 p->command = RXRPC_CMD_CHARGE_ACCEPT;
465 case RXRPC_EXCLUSIVE_CALL:
471 case RXRPC_UPGRADE_SERVICE:
477 case RXRPC_TX_LENGTH:
478 if (p->call.tx_total_len != -1 || len != sizeof(__s64))
480 p->call.tx_total_len = *(__s64 *)CMSG_DATA(cmsg);
481 if (p->call.tx_total_len < 0)
485 case RXRPC_SET_CALL_TIMEOUT:
486 if (len & 3 || len < 4 || len > 12)
488 memcpy(&p->call.timeouts, CMSG_DATA(cmsg), len);
489 p->call.nr_timeouts = len / 4;
490 if (p->call.timeouts.hard > INT_MAX / HZ)
492 if (p->call.nr_timeouts >= 2 && p->call.timeouts.idle > 60 * 60 * 1000)
494 if (p->call.nr_timeouts >= 3 && p->call.timeouts.normal > 60 * 60 * 1000)
505 if (p->call.tx_total_len != -1 && p->command != RXRPC_CMD_SEND_DATA)
512 * Create a new client call for sendmsg().
513 * - Called with the socket lock held, which it must release.
514 * - If it returns a call, the call's lock will need releasing by the caller.
516 static struct rxrpc_call *
517 rxrpc_new_client_call_for_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg,
518 struct rxrpc_send_params *p)
519 __releases(&rx->sk.sk_lock.slock)
520 __acquires(&call->user_mutex)
522 struct rxrpc_conn_parameters cp;
523 struct rxrpc_call *call;
526 DECLARE_SOCKADDR(struct sockaddr_rxrpc *, srx, msg->msg_name);
530 if (!msg->msg_name) {
531 release_sock(&rx->sk);
532 return ERR_PTR(-EDESTADDRREQ);
536 if (key && !rx->key->payload.data[0])
539 memset(&cp, 0, sizeof(cp));
540 cp.local = rx->local;
542 cp.security_level = rx->min_sec_level;
543 cp.exclusive = rx->exclusive | p->exclusive;
544 cp.upgrade = p->upgrade;
545 cp.service_id = srx->srx_service;
546 call = rxrpc_new_client_call(rx, &cp, srx, &p->call, GFP_KERNEL,
547 atomic_inc_return(&rxrpc_debug_id));
548 /* The socket is now unlocked */
550 rxrpc_put_peer(cp.peer, rxrpc_peer_put_discard_tmp);
551 _leave(" = %p\n", call);
556 * send a message forming part of a client call through an RxRPC socket
557 * - caller holds the socket locked
558 * - the socket may be either a client socket or a server socket
560 int rxrpc_do_sendmsg(struct rxrpc_sock *rx, struct msghdr *msg, size_t len)
561 __releases(&rx->sk.sk_lock.slock)
563 enum rxrpc_call_state state;
564 struct rxrpc_call *call;
565 unsigned long now, j;
566 bool dropped_lock = false;
569 struct rxrpc_send_params p = {
570 .call.tx_total_len = -1,
571 .call.user_call_ID = 0,
572 .call.nr_timeouts = 0,
573 .call.interruptibility = RXRPC_INTERRUPTIBLE,
575 .command = RXRPC_CMD_SEND_DATA,
582 ret = rxrpc_sendmsg_cmsg(msg, &p);
584 goto error_release_sock;
586 if (p.command == RXRPC_CMD_CHARGE_ACCEPT) {
588 if (rx->sk.sk_state != RXRPC_SERVER_LISTENING)
589 goto error_release_sock;
590 ret = rxrpc_user_charge_accept(rx, p.call.user_call_ID);
591 goto error_release_sock;
594 call = rxrpc_find_call_by_user_ID(rx, p.call.user_call_ID);
597 if (p.command != RXRPC_CMD_SEND_DATA)
598 goto error_release_sock;
599 call = rxrpc_new_client_call_for_sendmsg(rx, msg, &p);
600 /* The socket is now unlocked... */
602 return PTR_ERR(call);
603 /* ... and we have the call lock. */
605 if (READ_ONCE(call->state) == RXRPC_CALL_COMPLETE)
608 switch (READ_ONCE(call->state)) {
609 case RXRPC_CALL_UNINITIALISED:
610 case RXRPC_CALL_CLIENT_AWAIT_CONN:
611 case RXRPC_CALL_SERVER_PREALLOC:
612 case RXRPC_CALL_SERVER_SECURING:
613 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
615 goto error_release_sock;
620 ret = mutex_lock_interruptible(&call->user_mutex);
621 release_sock(&rx->sk);
627 if (p.call.tx_total_len != -1) {
629 if (call->tx_total_len != -1 ||
633 call->tx_total_len = p.call.tx_total_len;
637 switch (p.call.nr_timeouts) {
639 j = msecs_to_jiffies(p.call.timeouts.normal);
640 if (p.call.timeouts.normal > 0 && j == 0)
642 WRITE_ONCE(call->next_rx_timo, j);
645 j = msecs_to_jiffies(p.call.timeouts.idle);
646 if (p.call.timeouts.idle > 0 && j == 0)
648 WRITE_ONCE(call->next_req_timo, j);
651 if (p.call.timeouts.hard > 0) {
652 j = msecs_to_jiffies(p.call.timeouts.hard);
655 WRITE_ONCE(call->expect_term_by, j);
656 rxrpc_reduce_call_timer(call, j, now,
657 rxrpc_timer_set_for_hard);
662 state = READ_ONCE(call->state);
663 _debug("CALL %d USR %lx ST %d on CONN %p",
664 call->debug_id, call->user_call_ID, state, call->conn);
666 if (state >= RXRPC_CALL_COMPLETE) {
667 /* it's too late for this call */
669 } else if (p.command == RXRPC_CMD_SEND_ABORT) {
671 if (rxrpc_abort_call("CMD", call, 0, p.abort_code, -ECONNABORTED))
672 ret = rxrpc_send_abort_packet(call);
673 } else if (p.command != RXRPC_CMD_SEND_DATA) {
676 ret = rxrpc_send_data(rx, call, msg, len, NULL, &dropped_lock);
681 mutex_unlock(&call->user_mutex);
683 rxrpc_put_call(call, rxrpc_call_put_sendmsg);
684 _leave(" = %d", ret);
688 release_sock(&rx->sk);
693 * rxrpc_kernel_send_data - Allow a kernel service to send data on a call
694 * @sock: The socket the call is on
695 * @call: The call to send data through
696 * @msg: The data to send
697 * @len: The amount of data to send
698 * @notify_end_tx: Notification that the last packet is queued.
700 * Allow a kernel service to send data on a call. The call must be in an state
701 * appropriate to sending data. No control data should be supplied in @msg,
702 * nor should an address be supplied. MSG_MORE should be flagged if there's
703 * more data to come, otherwise this data will end the transmission phase.
705 int rxrpc_kernel_send_data(struct socket *sock, struct rxrpc_call *call,
706 struct msghdr *msg, size_t len,
707 rxrpc_notify_end_tx_t notify_end_tx)
709 bool dropped_lock = false;
712 _enter("{%d,%s},", call->debug_id, rxrpc_call_states[call->state]);
714 ASSERTCMP(msg->msg_name, ==, NULL);
715 ASSERTCMP(msg->msg_control, ==, NULL);
717 mutex_lock(&call->user_mutex);
719 _debug("CALL %d USR %lx ST %d on CONN %p",
720 call->debug_id, call->user_call_ID, call->state, call->conn);
722 switch (READ_ONCE(call->state)) {
723 case RXRPC_CALL_CLIENT_SEND_REQUEST:
724 case RXRPC_CALL_SERVER_ACK_REQUEST:
725 case RXRPC_CALL_SERVER_SEND_REPLY:
726 ret = rxrpc_send_data(rxrpc_sk(sock->sk), call, msg, len,
727 notify_end_tx, &dropped_lock);
729 case RXRPC_CALL_COMPLETE:
730 read_lock_bh(&call->state_lock);
732 read_unlock_bh(&call->state_lock);
735 /* Request phase complete for this client call */
736 trace_rxrpc_rx_eproto(call, 0, tracepoint_string("late_send"));
742 mutex_unlock(&call->user_mutex);
743 _leave(" = %d", ret);
746 EXPORT_SYMBOL(rxrpc_kernel_send_data);
749 * rxrpc_kernel_abort_call - Allow a kernel service to abort a call
750 * @sock: The socket the call is on
751 * @call: The call to be aborted
752 * @abort_code: The abort code to stick into the ABORT packet
753 * @error: Local error value
754 * @why: 3-char string indicating why.
756 * Allow a kernel service to abort a call, if it's still in an abortable state
757 * and return true if the call was aborted, false if it was already complete.
759 bool rxrpc_kernel_abort_call(struct socket *sock, struct rxrpc_call *call,
760 u32 abort_code, int error, const char *why)
764 _enter("{%d},%d,%d,%s", call->debug_id, abort_code, error, why);
766 mutex_lock(&call->user_mutex);
768 aborted = rxrpc_abort_call(why, call, 0, abort_code, error);
770 rxrpc_send_abort_packet(call);
772 mutex_unlock(&call->user_mutex);
775 EXPORT_SYMBOL(rxrpc_kernel_abort_call);
778 * rxrpc_kernel_set_tx_length - Set the total Tx length on a call
779 * @sock: The socket the call is on
780 * @call: The call to be informed
781 * @tx_total_len: The amount of data to be transmitted for this call
783 * Allow a kernel service to set the total transmit length on a call. This
784 * allows buffer-to-packet encrypt-and-copy to be performed.
786 * This function is primarily for use for setting the reply length since the
787 * request length can be set when beginning the call.
789 void rxrpc_kernel_set_tx_length(struct socket *sock, struct rxrpc_call *call,
792 WARN_ON(call->tx_total_len != -1);
793 call->tx_total_len = tx_total_len;
795 EXPORT_SYMBOL(rxrpc_kernel_set_tx_length);